APIMODELSAPIMODELS

Gemini 4 Carbon Apps, Real Codebases and Tooling Prompts

223 things people built, tested and compared with Gemini 4 Carbon in its first weeks: 3D scenes, games, simulations, films made from code, real bug fixes and research, and dozens of same-prompt tests against Claude Fable 5.2 and Gemini 4 Argon. Each is credited and linked to the original post, with the prompt where the creator published one.

← All 223 Gemini 4 Carbon prompts

Last updated: · Gemini 4 Carbon API — pricing & playground · API docs

TL;DR

Gemini 4 Carbon (model id gemini-4-carbon) is Google's deeper-reasoning Gemini 4 model. Google built it on Gemini 4 Argon and describes the difference as more thinking tokens per turn; it opened to vetted cybersecurity defenders on 28 October 2026 and to every paid Gemini API developer on 10 November. It reads text, images, video, audio and PDFs with a 1,048,576-token context and writes text, so the scenes, games and films here are code it wrote (Three.js, WebGL, Blender Python, CUDA C++, Web Audio), mostly run by Google's Antigravity agent. Of the 223 cases, 18 include the prompt as published, 33 only the creator's request and 172 no prompt; we never fill one in. In same-prompt tests creators usually found Claude Fable 5.2's results more polished and Carbon's cost a small fraction of Fable's. On apimodels.app Carbon costs $1.10 per 1M input tokens and $5.50 per 1M output through 30 November 2026 and $2.20 / $11 from 1 December, against Google's $2 / $10 and $4 / $20.

Source: Google · Gemini 4 Carbon (the model vendor’s official site)

Hover to preview · click for sound

Windows 11 running locally on a foldable iPhone

Prompt not published

No prompt published. With Carbon's help, the author got Windows 11, the actual desktop ARM OS, running locally on an iPhone Duo with touch controls. Unfolded, it works as a pocket PC that can even run Antigravity coding agents; a demo and repo are linked.

Clip:
Video, 60 s. About 115K views and 2,026 likes on X as of 12 November 2026.
Note:
A hardware and emulation hack; the post does not say which parts Carbon did.
video by @ammaarSource
Hover to preview · click for sound

New features for a floor plan generator

Prompt not published

No prompt published. A week after releasing a beta floor plan generator, the author used Carbon to add features: a shareable walkthrough mode, door and window generation based on rooms, a section cropping box, JSON export and import with LLM prompting, and an AR mode.

Clip:
Video, 60 s. About 4,256 views and 73 likes on X as of 12 November 2026.
Note:
Free to try at floorcraft.chirostudio.xyz (link in the post).
Hover to preview · click for sound

Shader fixes and QA for a Unity shadow-play video

Prompt not published

No prompt published. In the author's pipeline, MiniMax H3 Image refines holo images while Carbon handles the shader fixes and checks. In the quoted earlier post, Carbon refined a shadow-play video used in Unity, and the author says Argon's frequent misbehavior is gone.

Clip:
Video, 48 s. About 808 views and 56 likes on X as of 12 November 2026.
Note:
The main post is mostly about MiniMax H3 Image; Carbon's part is the shader work.
video by @ShadeLurkSource
Rerunning seven real business-code issues

Rerunning seven real business-code issues

Prompt not published

No prompt published. Once Carbon reached Antigravity subscribers, the author reran the 7 issues from an earlier real-business-code comparison of Codex, Grok, Kimi and Gemini 4, and saw a big improvement.

Clip:
Image. About 10K views and 27 likes on X as of 12 November 2026.
Tool:
Antigravity
Numbers the author reported:
7 个 Issue
Compared with:
Gemini 4 (earlier run), Codex, Grok, Kimi. Big improvement.
Note:
The quoted earlier post links the author's write-up of the original comparison.
video by @luoleiorgSource

Antigravity run that never edited a line of code

Prompt not published

No prompt published. On a US Google Ultra plan, Carbon High in Antigravity spent 40+ minutes checking a medium-complexity coding task without changing a line, then ended with "Agent execution terminated due to error."; an HTML demo also ignored instructions. The author's guess: "这种情况之前从未发生,疑似上下文爆炸了。"

Where:
V2EX, 2026-11-12
Tool:
Antigravity
Numbers the author reported:
40+ minutes without changing a line · about $10 of equivalent API cost (author's estimate)
Note:
A failure report. The cost is the author's estimate of equivalent API cost.
Watch on the original post
video by weiyunjunSource

44% prompt-cache hit rate for Factory-hosted gemini-4-carbon

Prompt not published

No prompt published. In a 4,392-message Droid session with 29 compactions on 2026-11-11, Factory-hosted gemini-4-carbon cached only 44.4% of input (184.3M uncached vs 147.0M cache-read tokens). The same model via BYOK the same day cached at 88.9%, and other hosted models at 96-99%.

Where:
GitHub, 2026-11-12
Tool:
Droid CLI 0.250.0 + Factory desktop, Max plan
Numbers the author reported:
184.3M uncached input vs 147.0M cache-read tokens (44.4%) · 4,392-message session with 29 compactions · BYOK cached at 88.9% · other hosted models 96-99%
Watch on the original post
video by marovoleSource

Patching an abandoned fishing game's binary for diamond fish

Prompt not published

No prompt published. With "Google Gemini Carbon", the author reverse-engineered the files of Call of the Wild: The Angler to make every water cell diamond-eligible, spawn only diamond fish and remove hook-size limits, something they had failed to do for 4 years. The full session transcript (.docx) and dozens of analysis scripts are in the repo.

Where:
GitHub, 2026-11-11
Numbers the author reported:
failed for 4 years before this
Note:
Harness not named. The full session transcript is published in the repo as a .docx.
Also:
https://news.ycombinator.com/item?id=50458062
Watch on the original post
video by gbolcerSource

Six merged production fixes replayed 20 times: Carbon vs Argon

Prompt not published

No prompt published. Six merged changes from a production monorepo were replayed on clean snapshots in the Antigravity CLI at medium effort and graded by the tests shipped with each fix. Carbon and Argon came out effectively tied, with Carbon using more output tokens.

Where:
Blog post, 2026-11-11
Tool:
Antigravity CLI
Numbers the author reported:
runs passing every test: Carbon 17/20 vs Argon 16/20 · new tests fixed: 96% vs 98% · Carbon about 30 min per fix (max 46) · output tokens 1.82M vs 1.37M · list cost $60.38 vs $57.68 · 4 parallel sessions on a $99.99 Google AI Ultra plan · Ultra quota stopped all four sessions after about 2 hours (13 runs finished, 4 died mid-fix)
Compared with:
Argon. Effectively a tie; in the author's words, Carbon is Argon allowed to spend more.
Note:
The author is not named on the page. Earlier Argon runs used Claude Code through a third-party API.
Watch on the original post
video by paddo.devSource

Long agentic coding sessions: Carbon vs Opus 5.5 vs Fable 5.2

Prompt not published

No prompt published. Long agentic coding sessions on production services, with the Claude models in Claude Code at medium effort, Carbon through Google AI Studio at default thinking, and a judge model with a fixed answer key. Carbon scored lowest overall and took longest, but cost the least per session.

Where:
Blog post, 2026-11-11
Tool:
Google AI Studio
Numbers the author reported:
overall: Opus 85.0%, Fable 83.1%, Carbon 79.5% · one-shot: Carbon 89.1% (Fable 89.4%) · cost per long session: Carbon $2.75, Opus $5.42, Fable $8.13 · time: 32 vs 17 vs 15 min · 20-turn session: Opus 89.4% ($43.37, 112 min) vs Carbon 79.4% ($21, 173 min)
Compared with:
Opus 5.5, Fable 5.2. Keep Claude Opus 5.5 as the default for long agentic coding.
Note:
Company blog; the byline is unclear.
Watch on the original post
video by BitoSource

Strands SDK integration tests run against gemini-4-carbon

Prompt not published

No prompt published. The Strands SDK's Python Gemini suite ran against gemini-4-carbon through Vertex AI: 23 passed and 3 failed (one SDK bug not specific to Carbon, one test-param artifact, and one quota 429 that passed on re-run).

Where:
GitHub, 2026-11-11
Tool:
Python Gemini suite through Vertex AI
Numbers the author reported:
23 passed, 3 failed
Note:
Posted by an AI agent account; the issue is AI-drafted.
Watch on the original post

MCP connector for Carbon design reviews from inside Codex

Prompt not published

No prompt published. An MCP connector that lets Codex ask Gemini 4 Carbon High for design reviews via Codex -> stdio MCP -> `agy acp` -> Antigravity, checking each session is `gemini-4-carbon` at `thought_level=high` before asking. Tested on Windows with Antigravity CLI 1.5.5.

Where:
GitHub, 2026-11-11
Tool:
Codex -> stdio MCP -> `agy acp` -> Antigravity
Numbers the author reported:
Antigravity CLI 1.5.5
Note:
The author warns that using a personal Antigravity subscription this way may break Antigravity's terms.
Watch on the original post
video by wpv-chanSource
Voice assistant from a PRD in 4.5 hours, plus bug fixesYouTube

Voice assistant from a PRD in 4.5 hours, plus bug fixes

Prompt not published

No prompt published. From a PRD, Carbon spent about 4.5 hours building a "Jarvis" voice assistant that clicks, types and fills forms on LiveKit Inference, testing it without a mic by generating speech and fixing the vision model's click offset itself. Carbon, Argon and Opus then fixed two real LiveKit Agents bugs and security-reviewed the homepage voice agent.

Where:
YouTube video, 2026-11-10
Numbers the author reported:
about 4.5 h build with no questions asked · security review real issues: Carbon 6, Argon 6, Opus 5, none invented
Compared with:
Argon, Opus 5.5. Opus was faster on both bug fixes; Carbon's top security finding was a framework bug the others missed, though it overrated severity.
Note:
Made with LiveKit. A judge LLM checked the security findings; the framework bug Carbon found was reported and fixed.
video by codeSTACKrSource

How to write great Gemini 4 Carbon prompts

  1. 1Write a spec, not a wish. Creators who gave Carbon one vague line mostly got average results; one who ran 15 projects summed it up as: don't make wishes at it. The striking results came from briefs that named the stack, the deliverable and what done looks like, up to a spec of about 10,000 characters for a three-minute violin concert film.
  2. 2Show it a reference. Carbon reads video. One creator gave it the gameplay video of a shooter Claude Fable 5.2 had built and asked for the same game or better; another gave it Anthropic's own Fable 5.2 demo video to rebuild as a web page.
  3. 3Make it look at its own output. The shooter rebuild ran a headless-Chrome screenshot loop that caught and fixed three bugs on its own, in 48 minutes for $3.82. Another creator had Carbon review its screenshots like a VFX supervisor four times, about two hours and $7.40, turning a block of voxels into a convincing dinosaur.
  4. 4Name the format. "One HTML file. No external assets." is half of the 5,000-drone light show prompt. Simon Willison had to add "build that into a standalone HTML file" to get a page he could open on its own.
  5. 5Split big builds across agents. The largest builds used Antigravity's multi-agent /swarm mode: nine agents and 22,910 lines for a diorama builder in 1 hour 41 minutes, and five waves of agent critique for an open-world game.
  6. 6Follow up. One creator posted the same island and castle after the first prompt and after three follow-ups; the fire went from nine orange cubes to something close to real. The gap, in the author's words, came from the follow-ups.
  7. 7Check what it says it finished. In one six-app test every one of Carbon's final summaries claimed at least one feature that turned out broken or overstated. Ask it to click through the result in a browser and report what it saw.

Prompt formula

Carbon build brief = the deliverable and format (one HTML file, an MP4, a repo) + the stack and constraints + what done looks like + a reference (an image or video it should match) + a self-check (open it in a browser, screenshot, fix) + a stop condition or time and token cap

Example (our template, not a creator's prompt): Build a browser game: a night-time arcade racer on a neon highway. Three.js from a CDN, music and sound in Web Audio, everything else in one index.html. Match the mood of the attached video. Done means: it loads with no console errors, steering and crashes work, and a lap takes about 90 seconds. Open it in a headless browser, take screenshots at the start and mid-lap, fix what looks wrong, then tell me what you checked. Do not ask questions.

Frequently asked questions

Can Gemini 4 Carbon make 3D scenes, games or videos by itself?

Not directly. Carbon takes text, images, video, audio and PDFs and returns text. Every scene, game and film here was rendered from code it wrote: Three.js and WebGL pages, Blender Python, CUDA or C++ programs, Web Audio for sound. In a chat window you get the code or an HTML file; an agent such as Antigravity, Cursor or GitHub Copilot can run, render and test it.

Which tool did people use with Carbon?

Mostly Google's Antigravity, often with its /swarm multi-agent mode. Others used Cursor, GitHub Copilot, AI Studio, Vertex AI or the API from their own harness. Where the creator said, the card names the tool.

Is Carbon better than Claude Fable 5.2 for this kind of work?

Usually not on polish, often yes on cost. In most same-prompt tests here Fable 5.2 produced the better-looking, more complete result, while Carbon cost a fraction as much and sometimes won outright: $7.54 against $41.71 for three jobs in one test, where Carbon won two of the three.

Is Carbon better than Gemini 4 Argon?

Mostly yes, at the cost of time and tokens. The two cost the same per token; in these tests Carbon usually produced the stronger result and took several times longer, for example 81 minutes against 14 in one Blender test that Argon had failed. Google still recommends Argon for most everyday coding.

How much did these builds cost?

As the creators reported: $0.45 for a one-shot 3D platformer of 2,599 lines, $3.82 for the shooter rebuild, about $7.40 for four rounds of self-review and $27.68 for three long builds. Cost depends on the thinking level and how long the agent runs; thinking tokens are billed as output.

How do I run Gemini 4 Carbon?

On apimodels.app, call model gemini-4-carbon on the OpenAI-compatible /v1/chat/completions endpoint or the native Gemini endpoint with one key, from anywhere including mainland China. Raise the output limit to at least 32,768 tokens, stream long answers and allow a 15-minute timeout, because Carbon thinks for a while before it writes.

Can I reuse these prompts and videos?

The works belong to their creators. Clips from X play from X's own servers, videos on YouTube and Bilibili play in their own players, and every card links to the original; we do not host or re-upload any of them. Published prompts are quoted with credit, in full when short and excerpted when long. If you made one of these and want the card removed or the credit corrected, email us.

Run these prompts on apimodels.app

Gemini 4 Carbon is live here alongside 160+ other image, video, audio and language models — one API, one key, pay as you go, with $0.10 of trial credit on sign-up.

This library is open source on GitHub

The whole collection is on GitHub as Markdown and JSON: all 223 cases with credits and links to the original posts and pages, the tools and numbers each creator reported, and a summary of the same-prompt tests. The videos and images stay with their creators and are linked, not re-hosted. Prompts up to 400 characters are quoted in full; longer ones are excerpted, and unpublished ones are left out. Our own write-ups are licensed CC BY 4.0. Open an issue to add a case or correct a credit.

View on GitHub

Other prompt libraries